From 97a38958bdcd0bff4059859998b189e4ed812730 Mon Sep 17 00:00:00 2001 From: Jin Qiao <89779290+JinBridger@users.noreply.github.com> Date: Thu, 2 Nov 2023 15:34:25 +0800 Subject: [PATCH] LLM: add CodeLlama CPU and GPU examples (#9338) * LLM: add codellama CPU pytorch examples * LLM: add codellama CPU transformers examples * LLM: add codellama GPU transformers examples * LLM: add codellama GPU pytorch examples * LLM: add codellama in readme * LLM: add LLaVA link --- README.md | 4 +- python/llm/README.md | 4 +- .../Model/codellama/README.md | 66 ++++++++++++++++ .../Model/codellama/generate.py | 67 ++++++++++++++++ .../PyTorch-Models/Model/codellama/README.md | 63 +++++++++++++++ .../Model/codellama/generate.py | 61 ++++++++++++++ .../Model/codellama/generate.py | 79 +++++++++++++++++++ .../Model/codellama/readme.md | 56 +++++++++++++ .../PyTorch-Models/Model/codellama/README.md | 58 ++++++++++++++ .../Model/codellama/generate.py | 75 ++++++++++++++++++ 10 files changed, 531 insertions(+), 2 deletions(-) create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/README.md create mode 100644 python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/generate.py create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/codellama/README.md create mode 100644 python/llm/example/CPU/PyTorch-Models/Model/codellama/generate.py create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/generate.py create mode 100644 python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/readme.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/codellama/README.md create mode 100644 python/llm/example/GPU/PyTorch-Models/Model/codellama/generate.py diff --git a/README.md b/README.md index e4c295c0df6..298413a7ed7 100644 --- a/README.md +++ b/README.md @@ -156,10 +156,12 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa | Phi-1_5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-1_5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-1_5) | | Flan-t5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/flan-t5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/flan-t5) | | Qwen-VL | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | | -| LLaVA | [link](python/llm/example/CPU/PyTorch-Models/Model/llava) | [link](python/llm/example/GPU/PyTorch-Models/Model/llava) | +| LLaVA | [link](python/llm/example/CPU/PyTorch-Models/Model/llava) | [link](python/llm/example/GPU/PyTorch-Models/Model/llava) | +| CodeLlama | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama) | | Skywork | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/skywork) | | + ***For more details, please refer to the `bigdl-llm` [Document](https://test-bigdl-llm.readthedocs.io/en/main/doc/LLM/index.html), [Readme](python/llm), [Tutorial](https://github.com/intel-analytics/bigdl-llm-tutorial) and [API Doc](https://bigdl.readthedocs.io/en/latest/doc/PythonAPI/LLM/index.html).*** --- diff --git a/python/llm/README.md b/python/llm/README.md index 58ae2ab9221..986a788a740 100644 --- a/python/llm/README.md +++ b/python/llm/README.md @@ -64,7 +64,9 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa | Flan-t5 | [link](example/CPU/HF-Transformers-AutoModels/Model/flan-t5) | [link](example/GPU/HF-Transformers-AutoModels/Model/flan-t5) | | Qwen-VL | [link](example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | | | LLaVA | [link](example/CPU/PyTorch-Models/Model/llava) | [link](example/GPU/PyTorch-Models/Model/llava) | -| Skywork | [link](example/CPU/HF-Transformers-AutoModels/Model/skywork) | | +| CodeLlama | [link](example/CPU/HF-Transformers-AutoModels/Model/codellama) | [link](example/GPU/HF-Transformers-AutoModels/Model/codellama) | +| Skywork | [link](example/CPU/HF-Transformers-AutoModels/Model/skywork) | | + ### Working with `bigdl-llm` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/README.md new file mode 100644 index 00000000000..8fc019576ab --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/README.md @@ -0,0 +1,66 @@ +# CodeLlama +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on CodeLlama models. For illustration purposes, we utilize the [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) as reference CodeLlama models. + +## 0. Requirements +To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a CodeLlama model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.9 +conda activate llm + +pip install bigdl-llm[all] # install bigdl-llm with 'all' option +``` + +### 2. Run +``` +python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the CodeLlama model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'codellama/CodeLlama-7b-hf'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +> **Note**: When loading the model in 4-bit, BigDL-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference. +> +> Please select the appropriate size of the CodeLlama model based on the capabilities of your machine. + +#### 2.1 Client +On client Windows machine, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py +``` + +#### 2.2 Server +For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket. + +E.g. on Linux, +```bash +# set BigDL-Nano env variables +source bigdl-nano-init + +# e.g. for a server with 48 cores per socket +export OMP_NUM_THREADS=48 +numactl -C 0-47 -m 0 python ./generate.py +``` + +#### 2.3 Sample Output +#### [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +def print_hello_world(): + +-------------------- Output -------------------- +def print_hello_world(): + print("Hello World") + + +def print_hello_world_with_name(name): + print("Hello " + name) + +``` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/generate.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/generate.py new file mode 100644 index 00000000000..3a383de796a --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/codellama/generate.py @@ -0,0 +1,67 @@ +# +# Copyright 2016 The BigDL Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import torch +import time +import argparse + +from bigdl.llm.transformers import AutoModelForCausalLM +from transformers import CodeLlamaTokenizer + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/docs/transformers/v4.34.1/model_doc/code_llama +CODELLAMA_PROMPT_FORMAT = "{prompt}\n" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeLlama model') + parser.add_argument('--repo-id-or-model-path', type=str, default="codellama/CodeLlama-7b-hf", + help='The huggingface repo id for the CodeLlama to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="def print_hello_world():", + help='Prompt to infer') + parser.add_argument('--n-predict', type=int, default=32, + help='Max tokens to predict') + + args = parser.parse_args() + model_path = args.repo_id_or_model_path + + # Load model in 4 bit, + # which convert the relevant layers in the model into INT4 format + model = AutoModelForCausalLM.from_pretrained(model_path, + load_in_4bit=True, + trust_remote_code=True) + + # Load tokenizer + tokenizer = CodeLlamaTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = CODELLAMA_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + st = time.time() + # if your selected model is capable of utilizing previous key/value attentions + # to enhance decoding speed, but has `"use_cache": false` in its model config, + # it is important to set `use_cache=True` explicitly in the `generate` function + # to obtain optimal performance with BigDL-LLM INT4 optimizations + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/CPU/PyTorch-Models/Model/codellama/README.md b/python/llm/example/CPU/PyTorch-Models/Model/codellama/README.md new file mode 100644 index 00000000000..b10746f17a7 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/codellama/README.md @@ -0,0 +1,63 @@ +# CodeLlama +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate CodeLlama models. For illustration purposes, we utilize the [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) as reference CodeLlama models. + +## Requirements +To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a CodeLlama model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations. +### 1. Install +We suggest using conda to manage the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#). + +After installing conda, create a Python environment for BigDL-LLM: +```bash +conda create -n llm python=3.9 # recommend to use Python 3.9 +conda activate llm + +pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option +``` + +### 2. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```powershell +python ./generate.py --prompt 'def print_hello_world():' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.2 Server +For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket. + +E.g. on Linux, +```bash +# set BigDL-Nano env variables +source bigdl-nano-init + +# e.g. for a server with 48 cores per socket +export OMP_NUM_THREADS=48 +numactl -C 0-47 -m 0 python ./generate.py --prompt 'def print_hello_world():' +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.3 Arguments Info +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the CodeLlama model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'codellama/CodeLlama-7b-hf'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### 2.3 Sample Output +#### [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) +```log +Inference time: xxxx s +-------------------- Output -------------------- +def print_hello_world(): + print("Hello World") + + +def print_hello_world_with_name(name): + print("Hello " + name) + +``` diff --git a/python/llm/example/CPU/PyTorch-Models/Model/codellama/generate.py b/python/llm/example/CPU/PyTorch-Models/Model/codellama/generate.py new file mode 100644 index 00000000000..f217676a461 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/codellama/generate.py @@ -0,0 +1,61 @@ +# +# Copyright 2016 The BigDL Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import torch +import time +import argparse + +from transformers import AutoModelForCausalLM, CodeLlamaTokenizer +from bigdl.llm import optimize_model + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/docs/transformers/v4.34.1/model_doc/code_llama +CODELLAMA_PROMPT_FORMAT = "{prompt}\n" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeLlama model') + parser.add_argument('--repo-id-or-model-path', type=str, default="codellama/CodeLlama-7b-hf", + help='The huggingface repo id for the CodeLlama to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="def print_hello_world():", + help='Prompt to infer') + parser.add_argument('--n-predict', type=int, default=32, + help='Max tokens to predict') + + args = parser.parse_args() + model_path = args.repo_id_or_model_path + + # Load model + model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True) + + # With only one line to enable BigDL-LLM optimization on model + model = optimize_model(model) + + # Load tokenizer + tokenizer = CodeLlamaTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = CODELLAMA_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt") + st = time.time() + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/generate.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/generate.py new file mode 100644 index 00000000000..4f215739965 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/generate.py @@ -0,0 +1,79 @@ +# +# Copyright 2016 The BigDL Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import torch +import intel_extension_for_pytorch as ipex +import time +import argparse + +from bigdl.llm.transformers import AutoModelForCausalLM +from transformers import CodeLlamaTokenizer + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/docs/transformers/v4.34.1/model_doc/code_llama +CODELLAMA_PROMPT_FORMAT = "{prompt}\n" + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeLlama model') + parser.add_argument('--repo-id-or-model-path', type=str, default="codellama/CodeLlama-7b-hf", + help='The huggingface repo id for the CodeLlama to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="def print_hello_world():", + help='Prompt to infer') + parser.add_argument('--n-predict', type=int, default=32, + help='Max tokens to predict') + + args = parser.parse_args() + model_path = args.repo_id_or_model_path + + # Load model in 4 bit, + # which convert the relevant layers in the model into INT4 format + model = AutoModelForCausalLM.from_pretrained(model_path, + load_in_4bit=True, + optimize_model=False, + trust_remote_code=True, + use_cache=True) + model = model.to('xpu') + + # Load tokenizer + tokenizer = CodeLlamaTokenizer.from_pretrained(model_path, + trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = CODELLAMA_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu') + + # ipex model needs a warmup, then inference time can be accurate + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + + # start inference + st = time.time() + # if your selected model is capable of utilizing previous key/value attentions + # to enhance decoding speed, but has `"use_cache": false` in its model config, + # it is important to set `use_cache=True` explicitly in the `generate` function + # to obtain optimal performance with BigDL-LLM INT4 optimizations + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + end = time.time() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Prompt', '-'*20) + print(prompt) + print('-'*20, 'Output', '-'*20) + print(output_str) diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/readme.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/readme.md new file mode 100644 index 00000000000..c9821a81077 --- /dev/null +++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/codellama/readme.md @@ -0,0 +1,56 @@ +# CodeLlama +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on CodeLlama models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) as a reference CodeLlama model. + +## 0. Requirements +To run these examples with BigDL-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for an CodeLlama model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel GPUs. +### 1. Install +We suggest using conda to manage environment: +```bash +conda create -n llm python=3.9 +conda activate llm +# below command will install intel_extension_for_pytorch==2.0.110+xpu as default +# you can install specific ipex/torch version for your need +pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu +``` + +### 2. Configures OneAPI environment variables +```bash +source /opt/intel/oneapi/setvars.sh +``` + +### 3. Run + +For optimal performance on Arc, it is recommended to set several environment variables. + +```bash +export USE_XETLA=OFF +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +``` + +``` +python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT +``` + +Arguments info: +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the CodeLlama model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'codellama/CodeLlama-7b-hf'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'def print_hello_world():'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### Sample Output +#### [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) +```log +Inference time: xxxx s +-------------------- Prompt -------------------- +def print_hello_world(): + +-------------------- Output -------------------- +def print_hello_world(): + print("Hello World!") + + +def print_hello_world_with_args(name): + print(f"Hello {name +``` diff --git a/python/llm/example/GPU/PyTorch-Models/Model/codellama/README.md b/python/llm/example/GPU/PyTorch-Models/Model/codellama/README.md new file mode 100644 index 00000000000..0b571551e9d --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/codellama/README.md @@ -0,0 +1,58 @@ +# CodeLlama +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate CodeLlama models. For illustration purposes, we utilize the [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) as reference CodeLlama models. + +## Requirements +To run these examples with BigDL-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information. + +## Example: Predict Tokens using `generate()` API +In the example [generate.py](./generate.py), we show a basic use case for a CodeLlama model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations on Intel GPUs. +### 1. Install +We suggest using conda to manage the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#). + +After installing conda, create a Python environment for BigDL-LLM: +```bash +conda create -n llm python=3.9 # recommend to use Python 3.9 +conda activate llm + +# below command will install intel_extension_for_pytorch==2.0.110+xpu as default +# you can install specific ipex/torch version for your need +pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu +``` + +### 2. Configures OneAPI environment variables +```bash +source /opt/intel/oneapi/setvars.sh +``` + +### 3. Run + +For optimal performance on Arc, it is recommended to set several environment variables. + +```bash +export USE_XETLA=OFF +export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 +``` + +```bash +python ./generate.py --prompt 'def print_hello_world():' +``` + +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the CodeLlama model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'codellama/CodeLlama-7b-hf'`. +- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `def print_hello_world():'`. +- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`. + +#### 2.3 Sample Output +#### [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) +```log +Inference time: xxxx s +-------------------- Output -------------------- +def print_hello_world(): + print("Hello World!") +print_hello_world() +def print_hello_world(): + print("Hello World!") + + +``` \ No newline at end of file diff --git a/python/llm/example/GPU/PyTorch-Models/Model/codellama/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/codellama/generate.py new file mode 100644 index 00000000000..065faadeb50 --- /dev/null +++ b/python/llm/example/GPU/PyTorch-Models/Model/codellama/generate.py @@ -0,0 +1,75 @@ +# +# Copyright 2016 The BigDL Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import torch +import intel_extension_for_pytorch as ipex +import time +import argparse + +from transformers import AutoModelForCausalLM, CodeLlamaTokenizer +from bigdl.llm import optimize_model + +# you could tune the prompt based on your own model, +# here the prompt tuning refers to https://huggingface.co/docs/transformers/v4.34.1/model_doc/code_llama +CODELLAMA_PROMPT_FORMAT = "{prompt}\n" + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeLlama model') + parser.add_argument('--repo-id-or-model-path', type=str, default="codellama/CodeLlama-7b-hf", + help='The huggingface repo id for the CodeLlama model to be downloaded' + ', or the path to the huggingface checkpoint folder') + parser.add_argument('--prompt', type=str, default="def print_hello_world():", + help='Prompt to infer') + parser.add_argument('--n-predict', type=int, default=32, + help='Max tokens to predict') + + args = parser.parse_args() + model_path = args.repo_id_or_model_path + + # Load model + model = AutoModelForCausalLM.from_pretrained(model_path, + trust_remote_code=True, + torch_dtype='auto', + low_cpu_mem_usage=True) + + # With only one line to enable BigDL-LLM optimization on model + model = optimize_model(model) + + model = model.to('xpu') + + # Load tokenizer + tokenizer = CodeLlamaTokenizer.from_pretrained(model_path, trust_remote_code=True) + + # Generate predicted tokens + with torch.inference_mode(): + prompt = CODELLAMA_PROMPT_FORMAT.format(prompt=args.prompt) + input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu') + # ipex model needs a warmup, then inference time can be accurate + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + + # start inference + st = time.time() + output = model.generate(input_ids, + max_new_tokens=args.n_predict) + torch.xpu.synchronize() + end = time.time() + output = output.cpu() + output_str = tokenizer.decode(output[0], skip_special_tokens=True) + print(f'Inference time: {end-st} s') + print('-'*20, 'Output', '-'*20) + print(output_str)